BMO off to promising start on its artificial intelligence projects

Register now

BMO Financial Group is already seeing signs that its investment in artificial intelligence is going to help it meet efficiency, loyalty and sales goals.

The Canadian banking company formed a new enterprise AI team in January that's built a customer-facing tool; two others have been developed by the bank's Digital Innovation program. All three are in the pilot phase and are expected to go live within the next six months.

BMO executives hope the new technology will help the company become more efficient. The idea is that it will allow customers to migrate from fully assisted channels like the branch and call center to digital channels like mobile, tablet and voice.

The technology is also meant to help grow digital sales and strengthen customer loyalty at the same time.

The team working on this initiative has dealt with many cultural, organizational and hiring challenges that banks tend to encounter when they try to deploy AI.

But the approach BMO devised — which is so holistic that the company refers to it as building an "AI organization" — is proving successful.

The three AI engines

The first problem BMO, which has 12 million customers, decided to tackle with AI was customers' cash shortages.

"Forty percent of our customers have trouble paying an unexpected $500 expense," said Peter Poon, managing director of digital product management and innovation at BMO. "Something like winter tires, a lot of people have trouble managing their cash flow to budget for it."

The new enterprise AI team built a system that mines customers' transaction histories and preauthorized payments to predict a cash flow shortfall 14 days in advance. In the pilot, these predictions have been right 95% of the time.

BMO has connected those insights with actions the customer can take. Customers can be automatically presented with a digital line of credit application with their data prepopulated. Or they can receive an authorized overdraft protection offer.

The second AI project created by Poon's group and the enterprise AI team is called BMO Quick Pay.

BMO focused on this because it found that 87% of Canadians pay at least one bill late per year.

To help customers deal with bills more efficiently, BMO used machine learning to create an automated bill payment tool.

Customers can take a picture of a bill or forward an e-statement to BMO. The software extracts the relevant data points from the bill (biller, amount, due date, etc.) and presents them to the customer over the mobile app or website. The customer types "pay" to confirm and the payment is made.

"Our customers are able to pay six times faster than the conventional bill payment process," Poon said. "On top of that, we've had well over 90% adoption amongst our pilot test base."

The third AI project in pilot is a conversational sales tool.

According to BMO, 5 million Canadians look for a better rewards credit card every year. This tool helps customers explore solutions and find the right card.

"Often customers find there's just so many choices," said Poon, who is based in Canada but has accountability for innovation, alerts and open banking for North America.

The software analyzes existing customer data and their responses to questions about what kinds of rewards they'd like. Then it makes a recommendation.

What is an AI organization?

The journey BMO went through to create what it refers to as an enterprise AI organization started with deciding on a strategy and hiring accordingly.

"AI is hard," said Yevgeniy Vahlis, head of the artificial intelligence technology group at BMO. "You have to think, what are you going to build internally versus what are you going to contract out, and consider the IT implications of that."

A lot of larger enterprises take existing analytics teams and upskill them, converting them to AI.

"Our approach is different," Vahlis said. "We built an AI team from scratch."

That involved seeking out a different pool of candidates than is typical for a bank, he said. "The majority of the people we brought in come from technology. These are academics, people who worked on self-driving cars or satellite image analysis. They don't know much about banking, but they know a lot about AI."

The usual career path for such people is that they complete their masters or Ph.D. program, then before anyone else gets a chance to recruit them, they're funneled directly into Google or Facebook, Vahlis noted dryly.

However, BMO has an advantage: Its Toronto and Montreal locations are AI hubs, and the company has built connections with the academic community in those cities.

Young people join the bank because they see opportunities, he said.

"It appeals to the ones that if they hadn't gone to grad school, they would have built a startup," he said. "The people who like the science but also get passionate about the product aspect of this: How do I build something?"

The next challenge is keeping these people happy in a bank.

"How do you keep them and how do you start bridging that cultural gap between the pure science/tech people and the finance and business culture?" Vahlis said.

"A lot of it has to do with creating the types of activities for these individuals that keep them engaged in the way with which they're familiar. It's a new world for them," he said. "They can be very comfortable with it as long as they get to solve hard scientific problems and they get to present their work and get credit for it."

Explaining AI and machine learning to the business people inside BMO is another challenge, especially since there are several data analytics groups in the company.

"The way we've been explaining and promoting AI capability within the bank is, AI brings the ability to consolidate multiple types of data and decisions in an automated way," Vahlis said. "So it's less about taking a particular data set and producing an insight. It's about taking in all the data you can, understanding it and then producing a multitude of insights from that understanding."

His group tries to build direct relationships with the business groups that use the AI products, insights and predictions his organization produces. The intent is to help the business people better understand and better monetize what the group is delivering.

Clear performance indicators are important for AI and businesspeople, Vahlis said.

"And it's important to hire the right leaders," he said. "They need to have technical expertise but also understand product and understand how to tie research work to products."

The AI team and Poon's group use agile delivery to pick up the pace of new tool development. Where previously two or three digital banking projects were completed each year, today there are sometimes as many as 18 projects in a release.

For reprint and licensing requests for this article, click here.